- Email: [email protected]
The gastroschisis prognostic score: reliable outcome prediction in gastroschisis
Journal of Pediatric Surgery (2012) 47, 1111–1117
www.elsevier.com/locate/jpedsurg
The gastroschisis prognostic score: reliable outcome prediction in gastroschisis☆,☆☆ Kyle N. Cowan a,⁎, Pramod S. Puligandla b,⁎, Jean-Martin Laberge b , Erik D. Skarsgard c , Sarah Bouchard d , Natalie Yanchar e , Peter Kim f , Shoo Lee g , Douglas McMillan e , Peter von Dadelszen c The Canadian Pediatric Surgery Network a
Children's Hospital of Eastern Ontario, Ottawa, Ontario, Canada K1H 8L1 Montreal Children's Hospital, Montreal, Quebec, Canada H3H 1P3 c British Columbia Children's Hospital, Vancouver, British Columbia, Canada V6H 3V4 d Hopital Ste-Justine, Montreal, Quebec, Canada H3T 1C5 e IWK Health Centre, Halifax, Nova Scotia, Canada B3K 6R8 f Hospital for Sick Children, Toronto, Ontario, Canada M5G 1X8 g Mt. Sinai Hospital, Toronto, Ontario, Canada M5G 1X5 b
Received 20 February 2012; accepted 5 March 2012
Key words: Gastroschisis; Bowel score; Outcomes; Prediction; Multivariate analysis; Validation; Risk stratification
Abstract Background/Purpose: Disease-specific outcome predictors are required for gastroschisis. We derived and validated a gastroschisis prognostic score (GPS) based on bowel appearance after birth. Methods: Visual scoring of bowel matting, necrosis, atresia, and perforation generated a novel gastroschisis bowel injury score recorded in a national database. Reweighting of score components by regression analysis led to assessments of model calibration and goodness of fit. The GPS was validated in subsequent cases. Results: Records from 225 infants were used for model derivation. Only intestinal necrosis independently predicted mortality by regression analysis (odds ratio, 11.5; 95% confidence interval, 4.2-31.4). Model recalibration identified that a GPS of 4 or more predicted mortality in 75% of nonsurvivors and 99% of survivors (P = .0001). A GPS of 2 or more demonstrated significantly worse survival outcomes compared with scores of 0 or 1 (length of stay: P = .011, days to first enteral feed: P = .013, days on total parenteral nutrition: P = .006). Model validation with 184 new patients yielded continued high-quality discrimination of outcomes. The GPS demonstrated “near-perfect” interobserver reliability between 2 surgeons (κ ≥ 0.86).
Abbreviations: GPS, gastroschisis prognostic score; CAPSNet, Canadian Pediatric Surgery Network; DTPNs, days on total parenteral nutrition; DFEFs, days to first enteral feed; LOS, length of stay; SNAP-II, Score for Neonatal Acute Physiology version II. ☆
Financial disclosure: None. Conflicts of interest: None. ⁎ Corresponding author. Tel.: +1 514 412 4438; fax: +1 514 412 4289. E-mail address: [email protected] (P.S. Puligandla). ☆☆
0022-3468/$ – see front matter © 2012 Elsevier Inc. All rights reserved. doi:10.1016/j.jpedsurg.2012.03.010
1112
K.N. Cowan et al. Conclusions: The GPS allows the accurate and reliable identification of high-risk groups for mortality and morbidity based on bowel appearance at birth. This information can drive discussions regarding family counseling, resource allocation, and new therapies for these patients. © 2012 Elsevier Inc. All rights reserved.
Gastroschisis is a congenital anomaly in which the bowel and other abdominal contents herniate through a paraumbilical, anterior abdominal wall defect. Although most of these infants survive, many experience significant gastrointestinal dysfunction [1]. Because the diagnosis of gastroschisis is made antenatally in most cases, attempts have been made to correlate antenatal ultrasonography findings with clinical outcome. Despite initial promise [2], variables such as bowel thickness and bowel dilatation on antenatal ultrasound have not been shown to be reliable predictors of morbidity or mortality in recent studies [3-6]. Postnatal variables such as prematurity and low birth weight have shown better correlation [7-9]. Furthermore, the classification of infants as complex, that is, those with intestinal atresia, stenosis, perforation, necrosis, or volvulus, has also been useful in identifying those infants at higher risk for morbidity and mortality [9-11]. These analyses, however, are often limited to single institutions and, therefore, may reflect small case numbers collected over long periods, often without the standardization of risk or outcome variables. The Canadian Pediatric Surgery Network (CAPSNet) is a multidisciplinary group from 16 tertiary perinatal centers that collects prospective, population-based data on gastroschisis and congenital diaphragmatic hernia. A major objective of CAPSNet has been the development of prognostication tools for gastroschisis. To date, no study has attempted to correlate the gross appearance of the bowel after birth with clinical outcome. For this purpose, we developed a standardized, visually determined, gastroschisis bowel injury scoring tool. The aim of this study was to use the prospective CAPSNet database to evaluate the outcome of infants with gastroschisis using a novel bowel injury scoring tool.
1. Patients and methods 1.1. Study population We performed a retrospective cohort study in which cases of gastroschisis born between May 1, 2005, and June 30, 2008, were abstracted from the CAPSNet database (inception cohort—cohort 1). This included data from diagnosis (usually prenatal) to death or discharge from the participating CAPSNet center. After initial analysis and modeling, study findings were prospectively validated in a new cohort of gastroschisis cases born between July 1, 2008, and October 31, 2010 (validation cohort—cohort 2).
1.2. Data collection Data were collected as previously described [4]. Briefly, a trained research assistant at each participating CAPSNet center abstracted prenatal and postnatal data using a customized data entry program and a standardized manual of operations and definitions with built-in error checking. The coded data, stripped of patient identifiers, was then transmitted electronically to a centralized, secure database for cleaning and storing. This process was overseen by a study coordinator and a multidisciplinary, geographically representative, steering committee consisting of pediatric surgeons, a neonatologist, a maternal-fetal medicine specialist, and an epidemiologist.
1.3. Gastroschisis bowel injury score To readily evaluate bowel injury in patients with gastroschisis and to standardize its assessment, a novel bowel injury scoring tool was developed. The composite score was based on key features of intestinal injury noted by the treating surgeon within 6 hours of birth. The score's component variables included the following: (a) bowel matting (none, 0; mild, 1; or severe, 2), (b) bowel necrosis (absent, 0; focal, 1; or diffuse, 2), (c) bowel atresia (absent, 0; suspected, 1, or present, 2), and (d) bowel perforation (absent, 0, or present, 2). Consensus-based, photographic examples with detailed, written descriptions of each quantitative bowel injury attribute were posted on the Web site (www.capsnetwork.org) in an attempt to standardize bowel scoring among surgeons. The injury score assessment was performed at the time of surgical consultation and prospectively recorded with the time of assessment.
1.4. Data analysis Data for analysis were abstracted directly from the CAPSNet database and analyzed with SPSS 12 for Windows statistical software (SPSS, Chicago, IL) using the tests subsequently described. The data parameters collected included demographic and perinatal risk variables (birth weight, gestational age, neonatal illness severity score [Score for Neonatal Acute Physiology version II, or SNAP-II] [12], and the gastroschisis bowel injury score). Mortality and nonmortality outcomes were analyzed, including length of stay (LOS), days to first enteral feed (DFEFs), days on total parenteral nutrition (DTPNs), and complications. The recorded complications included abdominal compartment
The gastroschisis prognostic score syndrome requiring treatment, culture-proven bacteremia or surgical site infection, a need for any reoperation, and the presence of severe TPN cholestasis at discharge (conjugated bilirubin N80 μmol/L). Univariate analyses for the comparison of risk-stratified groups were conducted using χ2 tests, whereas the MannWhitney test was performed for nonparametric variables without normal distribution. Multivariate predictors of mortality were analyzed with logistic regression where the conformity between actual and predicted outcome was assessed by model calibration and “goodness of fit” tests. To this end, the Omnibus test was used to assess the overall model fit, with a P value less than .05 confirming that the overall model parameters predicted mortality [13]. The Hosmer-Lemeshow technique was used to assess the goodness of fit, or the presence of a difference between the actual and modeled outcomes, where a significant model fit produced a P N .05 [14]. The individual components of the gastroschisis bowel injury scoring tool (matting, perforation, atresia, necrosis), as well as other putative outcome predictors identified in the literature, including gestational age, birth weight, and SNAPII [7,8,15], were subject to univariate analysis to identify those variables that were predictive of mortality. Variables displaying continued outcome prediction by logistic regression analyses were subsequently reweighted to derive a new prognostic score, denoted the gastroschisis prognostic score (GPS). The GPS was then used in all subsequent analyses including the risk stratification for mortality and the morbidity outcomes described previously. The validation of the GPS as an outcome prediction tool was subsequently performed on the next cohort of patients entered into the database from July 2008 to October 2010. Interobserver variability for the GPS was assessed using a second surgeon (blinded to the first surgeon's score) who provided a separate bowel scoring assessment based on a convenience sample of patients from the combined cohort. After biostatistical consultation, it was deemed that a convenience sample of more than 20% of the combined cohort was sufficient to analyze interobserver variability using Cohen κ statistic [16,17]. All secondary assessments for this analysis occurred between 6 and 24 hours of birth. All data analysis, methods, and statistical assessments were reviewed and overseen by biostatisticians and clinical research specialists from the University of Western Ontario (London, Ontario, Canada) and McGill University (Montreal, Quebec, Canada).
2. Results From a total of 510 gastroschisis patients entered into the CAPSNet database between May 2005 and October 2010, 409 patients with complete bowel scoring data were used for this study. There were no differences in demographic data
1113 including gestational age, birth weight, and SNAP-II score, nor mortality or survival outcomes between these patients and the 101 patients for whom complete bowel scoring data were not available (data not shown). Cohort 1 (inception cohort) consisted of 225 patients accrued from May 2005 to June 2008, whereas cohort 2 (validation cohort) consisted of the next 184 patients accrued from July 2008 to October 2010. The demographic details for the 2 cohorts are presented in Table 1. The overall mean gestational age and birth weight for the combined patient population were 36 weeks and 2526 g, respectively. The mean SNAP-II score for these patients was 8.4 ± 11.7 (median, 5; range, 0-64), suggesting that these infants had mild acute physiologic derangements and illness by this measure [12,18]. Ninetyone percent of all infants registered in the CAPSNet database during this time were evaluated with the bowel score within 6 hours of birth and exhibited a mean composite gastroschisis bowel injury score of 1 ± 1 (median, 1; range, 0-8). The overall survival rate for these patients was 96.8%. The median LOS was 37 days (range, 0-400 days), whereas the median DFEFs and median DTPNs were 14 (range, 1-78) and 28 (range, 4-309) days, respectively. All complications, when taken together, occurred at an average rate of 0.6 ± 0.8 events per patient (median, 0; range, 0-5). First, a predictive outcome model for mortality and survival outcomes was derived using bowel scoring (see “Methods”—“Gastroschisis bowel injury score”) and outcomes data for cohort 1. Multivariate logistic regression using only the 4 components of the gastroschisis bowel injury score demonstrated a robust model for the prediction of mortality in cohort 1. Univariate assessment of the individual contributions of each of the score components, as well as other potential predictor variables, identified that matting, necrosis, gestational age, and SNAP were variables that either had a greater individual contribution or could potentially contribute to the prediction model. Thus, these variables were candidates for either reweighting and/or Table 1 CAPSNet demographics and selected outcomes of patients with gastroschisis Parameter
Cohort 1 (n = 225)
Cohort 2 (n = 184)
Male, n (%) Gestational age (wk) Birth weight (g) SNAP-II b6 h to bowel score assessment, n (%) Gastroschisis bowel injury score Survival, n (%) LOS (median) (d) DFEFs (median) DTPNs (median) Complications
126 (56.0) 36 ± 2.2 2512 ± 583 8.5 ± 11.8 202 (89.8) 1±1 217 (96.4) 37 (1-349) 14 (2-78) 28 (5-221) 0.5 ± 0.8
96 (52.2) 36 ± 1.5 2543 ± 482 8.3 ± 11.8 172 (93.5) 1±1 179 (97.3) 36 (0-400) 14 (1-75) 28 (4-309) 0.6 ± 0.8
Data are presented as either mean ± SD or median (range) values, unless otherwise indicated.
1114
K.N. Cowan et al.
addition to the preliminary model that consisted of only the 4 components of the bowel injury score (Table 2). Importantly, univariate analysis identified necrosis, either focal or diffuse, as the only variable with the bowel injury score able to independently predict mortality (P = .0001) with an odds ratio of 11.5 (95% confidence interval, 4.2-31.4), whereas patients with matting, mild and severe assessed in aggregate, showed a trend toward significance (P = .055). Next, the individual components of the gastroschisis bowel injury score were reweighted within the prediction model based on the multiple logistic regression analyses (data not shown). Ultimately, the bowel injury score was reweighted in favor of both necrosis (absent, 0, or present, 4 [focal or diffuse]) and severe matting (none, 0; mild, 1; or severe, 4), thereby deriving a new prognostic score, denoted the GPS (Table 3). This revision directly reflected the independent predictive ability of any type of necrosis (ie, focal or diffuse) as well as the potential predictive ability of advanced (“severe”) matting. When these reweighted elements were combined into a mortality prediction model, the correct outcome was predicted in 99% (215/217) of survivors and 75% (6/8) of nonsurvivors. Furthermore, the Omnibus (P = .0001) and Hosmer-Lemeshow tests (P = .96) confirmed excellent model calibration and discrimination (Table 4). Interestingly, multivariate regression using the gastroschisis bowel injury score components with the addition of either gestational age or birth weight did not improve correlation with outcome, whereas the addition of the SNAP-II score, in fact, weakened the model fit (data not shown). The GPS permitted the identification of patients at a higher risk for death, with infants having a GPS of 4 or more (n = 38) experiencing a mortality rate of 16% (Table 4). When the patients with a GPS less than 4 (n = 187) were compared to this high-risk group, a statistically and clinically significant increase in survival (99%) was noted. Consistent with the univariate analysis, a significant difference in gestational age was observed between high- and low-risk groups, with higher gestational ages being associated with survival, whereas none was present when comparing groups on the basis of either birth weight or SNAP-II (data not shown). Table 2 Mortality outcome prediction for the components of the gastroschisis bowel injury score and other potential mortality predictor variables Parameter
P
Odds ratio
95% CI
Atresia Perforation Matting Necrosis Gestational age Birth weight SNAP-II
.108 .121 .055 .0001 .051 .179 .073
2.12 2.46 1.35 11.47 0.80 0.99 1.04
0.85-5.32 0.79-7.68 0.50-3.63 4.20-31.35 0.63-1.00 0.99-1.00 0.99-1.09
Abbreviation: CI, confidence interval.
Table 3 the GPS
Weighting “severe matting” and “necrosis” to derive
Matting Atresia Perforation Necrosis
None (0) Absent (0) Absent (0) Absent (0)
Mild (1) Suspected (1)
Severe (4) Present (2) Present (2) Present (4)
Based on logistic regression analyses, severe matting and necrosis were reweighted to acknowledge their individual contributions to the GPS. These criteria were initially given a maximum score of 2 in the preliminary model.
To be clinically relevant as a prognostic tool, it was important for the GPS to discriminate mortality from nonmortality outcomes, and for this reason, a separate logistic regression derivation of morbidity predictors was not performed. Rather, we used the same weighted scoring of bowel injury features that were used to predict mortality and performed a sequential analysis of composite GPS values (≥1, 2, 3, etc). This process identified a GPS of 2 or more as the threshold for the prediction of morbidity outcomes in cohort 1 (n = 51) (Table 5). Although there was no significant difference in the complication rates for patients with a GPS of 2 or more (n = 51) and those with a GPS less than 2 (n = 174), patients with a GPS of 2 or more demonstrated both a statistically and clinically significant increase in morbidity with respect to median LOS (63 vs 36, P = .011), DFEFs (19 vs 14, P = .013), and DTPNs (41 vs 27, P = .006). Next, model validation was undertaken using cohort 2. The GPS performed well, with excellent prediction of mortality in the validation cohort (Table 4). Moreover, the comparison of adverse survival outcomes between patients with a GPS of 2 or more and those with a GPS less than 2 demonstrated significant differences in all morbidities compared, including the rate of complications, an outcome that had only shown a trend toward significance in the inception cohort (cohort 1) (Table 5). Similar differences in mortality and adverse survival outcomes were observed Table 4 Risk-stratified mortality outcome prediction by the GPS in the derivation (cohort 1) and validation (cohort 2) patient groups Parameter
GPS ≥ 4
GPS b 4
Cohort 1 (n = 225) Mortality, n (%) Prediction of correct outcome (%) Omnibus test (P value) Hosmer-Lemeshow (P value) Cohort 2 (n = 184) Mortality, n (%) Prediction of correct outcome (%) Omnibus test (P value) Hosmer-Lemeshow (P value)
n = 38 6 (15.8) ⁎ 97.3 .0001 .96 n = 38 4 (10.5) † 97.3 .001 .921
n = 187 2 (1.1)
⁎ P = .0001. † P = .001.
n = 146 1 (0.7)
The gastroschisis prognostic score
1115
Table 5 Risk-stratified morbidity outcome prediction by the GPS in the derivation (cohort 1) and validation (cohort 2) patient groups Parameter
GPS ≥ 2
GPS b 2
Cohort 1 (n = 225) LOS (median) (d) DFEFs (median) DTPNs (median) Complications Cohort 2 (n = 184) LOS (median) (d) DFEFs (median) DTPNs (median) Complications
n = 51 63 (18-349) 19 (2-71) 41 (12-221) 0.8 ± 1.1 n = 49 68 (18-392) 19 (1-62) 50 (16-301) 0.9 ± 0.9
n = 174 36 (1-256) 14 (3-78) 27 (5-172) 0.5 ± 0.7 n = 135 33 (0-400) 13 (2-75) 27 (4-309) 0.5 ± 0.7
P .011 .013 .006 .06 .01 .002 .013 .003
Data are presented as either mean ± SD or median (range) values.
when both cohorts were examined in aggregate, where a significant difference in complication rate was again observed (data not shown). The interobserver variability in bowel injury scoring by the treating surgeon was evaluated using a second, blinded surgeon's score. This second evaluation occurred between 6 and 24 hours of birth and was applied to a convenience sample of 131 patients (32%) within the combined cohort. “Near perfect agreement” between observers was noted in the identification of patients at increased risk for mortality (GPS ≥ 4) and those at increased risk of morbidity (GPS ≥ 2), with κ values of 0.86 and 0.91, respectively [16,17].
3. Discussion To date, the prognostication of outcomes in infants with gastroschisis has remained controversial, and in light of the considerable resources used by these patients [19], there is much to be gained from a better understanding and, possibly, prediction of their clinical course [11]. To this end, many different antenatal and postnatal variables have been investigated [2,3,5,6,10,11]. This is the first study in the literature, however, to examine and correlate bowel appearance at birth with clinical outcome in the largest, populationbased group of infants with gastroschisis currently in the literature. This validated tool enables the identification of gastroschisis patients that are at high risk for mortality and morbidity in the immediate postnatal period. Risk stratification is not a new concept in gastroschisis. Caniano et al [10] and Molik et al [11] both published reports on the need to categorize infants with gastroschisis as either high or low risk based on several characteristics including the presence of intestinal atresia, stenosis, perforation, necrosis, or volvulus. These criteria continue to be used to identify high-risk patients [6,9], and some of these findings form the critical elements of the GPS. However, these risk stratification systems lack validation, and their dependence on the
diagnosis of intestinal atresia or stenosis (which may be delayed) limits their predictive use in the immediate newborn period. By contrast, the GPS is a rapid, efficient, and accurate means of risk categorization with excellent reproducibility that can be completed at the bedside without the need for monitoring or laboratory data. It can be performed within hours of birth using a simple, standardized, Web-based scoring tool that evaluates key attributes of bowel injury, specifically intestinal matting, atresia, perforation, and necrosis. A composite score of 2 or more heralds an infant at greater risk for multifactorial morbidity and prolonged hospitalization. A score of 4 or more identifies an infant with an additional higher likelihood of mortality. Although necrosis independently predicts mortality, severe bowel matting or a combination of bowel perforation, suspected atresia, and/or mild matting also leads to patient stratification into either of the increased risk groups. In this way, outcome prognostication can be performed shortly after birth to provide reliable and clinically relevant information that can be used to counsel families and select babies for innovative therapies, based on the anticipation of a complicated postnatal course, as well as providing a common benchmark upon which to compare future therapeutic interventions. Recently, the CAPSNet database used SNAP-II scoring to predict mortality in patients with gastroschisis [15]. This assessment, however, still required 12 hours of intensive hemodynamic, respiratory, and laboratory surveillance to calculate. Owen et al [20], in a population-based cohort of 393 infants with gastroschisis from the United Kingdom, avoided using the SNAP-II score in the risk stratification of patients because of the “burden” of data collection associated with this method. They simply defined complex patients as those with “additional bowel damage” such as perforation or atresia. In addition to SNAP-II, additional indicators of gestational maturity and fetal growth have also been shown to have a negative impact on LOS and feeding parameters in patients with gastroschisis [7,8]. Interestingly, mortality prediction was not improved by combinations of gestational age, birth weight, and the SNAP-II score with the GPS vs the GPS alone. This finding may reflect the failure of these individual variables to enhance the predictive ability already encompassed by the GPS, a potentially stronger predictor of outcome. An examination of the individual risk attributes of SNAP-II suggests that its predictive ability is focused on the uncommon clinical scenario of the physiologically unstable infant with gastroschisis [15]. Indeed, the risk of mortality was increased only for those infants with SNAP-II scores greater than 28, a value far exceeding the average SNAP-II score of 8 observed in the patient population with gastroschisis used in this study [15]. The GPS, by contrast, is, thus, a more useful predictor in its ability to accurately discriminate outcome in the more typical context of infants who are stable from a cardiorespiratory perspective, yet have a bowel injury of variable severity. Because the GPS has superior performance in this broader patient cohort, its use is more universal.
1116 Although the GPS was found to be a robust predictor of mortality outcomes based on Omnibus and HosmerLemeshow testing, it may still have limitations. The overall mortality rate in this cohort was very low, with only 13 deaths noted in the entire patient population. With such a low event rate, the model was only able to predict 77% (10/13) of these deaths. It is possible that with a considerably larger data set, the model's ability to predict mortality could improve. Furthermore, our results only reflect survival to discharge from hospital because we lack long-term follow-up data for these patients, and it is conceivable that patients with a high GPS could have died after discharge. Thus, given the very high overall survival rate in gastroschisis, accurate mortality prediction appears to be considerably less important than the prediction of adverse survival outcomes. Perhaps, the greatest use of the GPS is its ability to predict clinically relevant outcomes such as LOS, DFEFs, DTPNs, and complication rates. Therefore, we sought to establish risk thresholds that would discriminate between mortality and nonmortality outcomes. Although a separate model derivation for morbidities was considered, the difficulty in selecting the most “clinically important” morbidity outcome, as well as the questionable use of a tool that requires separate scoring systems for morbidity and mortality prediction, caused us to use a single tool (initially derived against the outcome of mortality), with different “threshold” composite scores that provided excellent discrimination of outcomes. Because a GPS less than 2 predicted a “low risk” of adverse outcome and a GPS of 4 or more predicted a “high risk” of adverse outcome (including mortality), we wondered if an independent “intermediate risk” group might also exist. However, the numbers of intermediate-risk patients (ie, GPS of 2 or 3) in isolation were comparatively small and did not provide as clear a separation of outcomes as we had hoped (data not shown). It is very possible that ongoing patient recruitment and the addition of long-term outcome data could enable us to better define an intermediate-risk group for both short- and long-term morbidity. Nonetheless, the value and use of the GPS rest in its ability to provide rapid and clinically relevant prognostic information on morbidity that can be used to drive important discussions regarding family counseling, institutional resource allocation, and the use of novel therapies for this population of patients.
4. Conclusion The GPS is an easy, efficient, reproducible, robust, and sensitive method for outcome prediction in neonates with gastroschisis. Unique, clinically relevant risk groups for mortality and morbidity can be identified using this score within hours of their birth. This information can then assist in the planning of multidisciplinary management regimens for these patients while informing discussions with parents related to outcome expectation.
K.N. Cowan et al.
Acknowledgments This work was supported by the Canadian Institute of Health Research Team in Maternal-Infant Care Grant MOP69050. We are indebted to Dr Jamie Seabrook and Dr Elise Mok of the Children's Health Research Institute, University of Western Ontario, and The Montreal Children's Hospital Research Institute, McGill University, respectively, for their statistical support throughout the course of the study. We would also like to thank Jennifer Claydon and Alana Gaumont for their superb management of the CAPSNet database. For their contributions to the preparation of this manuscript, we would also like to recognize Drs Andreana Bütter and Robert Baird.
References [1] Langer JC, Longaker MT, Crombleholme TM, et al. Etiology of intestinal damage in gastroschisis. I: effects of amniotic fluid exposure and bowel constriction in a fetal lamb model. J Pediatr Surg 1989;24: 992-7. [2] Bond SJ, Harrison MR, Filly RA, et al. Severity of intestinal damage in gastroschisis: correlation with prenatal sonographic findings. J Pediatr Surg 1988;23:520-5. [3] Piper HG, Jaksic T. The impact of prenatal bowel dilation on clinical outcomes in neonates with gastroschisis. J Pediatr Surg 2006;41: 897-900. [4] Skarsgard ED, Claydon J, Bouchard S, et al. Canadian Pediatric Surgical Network: a population-based pediatric surgery network and database for analyzing surgical birth defects. The first 100 cases of gastroschisis. J Pediatr Surg 2008;43:30-4 [discussion 34]. [5] Badillo AT, Hedrick HL, Wilson RD, et al. Prenatal ultrasonographic gastrointestinal abnormalities in fetuses with gastroschisis do not correlate with postnatal outcomes. J Pediatr Surg 2008;43:647-53. [6] Davis RP, Treadwell MC, Drongowski RA, et al. Risk stratification in gastroschisis: can prenatal evaluation or early postnatal factors predict outcome? Pediatr Surg Int 2009;25:319-25. [7] Ergun O, Barksdale E, Ergun FS, et al. The timing of delivery of infants with gastroschisis influences outcome. J Pediatr Surg 2005;40: 424-8. [8] Puligandla PS, Janvier A, Flageole H, et al. The significance of intrauterine growth restriction is different from prematurity for the outcome of infants with gastroschisis. J Pediatr Surg 2004;39:1200-4. [9] Payne NR, Pfleghaar K, Assel B, et al. Predicting the outcome of newborns with gastroschisis. J Pediatr Surg 2009;44:918-23. [10] Caniano DA, Brokaw B, Ginn-Pease ME. An individualized approach to the management of gastroschisis. J Pediatr Surg 1990;25:297-300. [11] Molik KA, Gingalewski CA, West KW, et al. Gastroschisis: a plea for risk categorization. J Pediatr Surg 2001;36:51-5. [12] Richardson DK, Corcoran JD, Escobar GJ, et al. SNAP-II and SNAPPE-II: simplified newborn illness severity and mortality risk scores. J Pediatr 2001;138:92-100. [13] Pearson ES, D'Agostino RB, Bowman KO. Test for departure from normality. Biometrika 1977;64:231-46. [14] Lemeshow S, Hosmer Jr DW. A review of goodness of fit statistics for use in the development of logistic regression models. Am J Epidemiol 1982;115:92-106. [15] Mills J, MacNab YC, Skarsgard ED, et al. Perinatal predictors of outcome in gastroschisis. J Perinatol 2010;30:809-13. [16] Cohen J. A coefficient of agreement for nominal scales. Educ Psychol Meas 1960;20:37-46.
The gastroschisis prognostic score [17] Viera AJ, Garrett JM. Understanding interobserver agreement: the kappa statistic. Fam Med 2005;37:360-3. [18] Chien LY, Whyte R, Thiessen P, et al. SNAP-II predicts severe intraventricular hemorrhage and chronic lung disease in the neonatal intensive care unit. J Perinatol 2002;22:26-30.
1117 [19] Sydorak RM, Nijagal A, Sbragia L, et al. Gastroschisis: small hole, big cost. J Pediatr Surg 2002;37:1669-72. [20] Owen A, Marven S, Johnson P, et al. Gastroschisis: a national cohort study to describe contemporary surgical strategies and outcomes. J Pediatr Surg 2010;45:1808-16.
www.elsevier.com/locate/jpedsurg
The gastroschisis prognostic score: reliable outcome prediction in gastroschisis☆,☆☆ Kyle N. Cowan a,⁎, Pramod S. Puligandla b,⁎, Jean-Martin Laberge b , Erik D. Skarsgard c , Sarah Bouchard d , Natalie Yanchar e , Peter Kim f , Shoo Lee g , Douglas McMillan e , Peter von Dadelszen c The Canadian Pediatric Surgery Network a
Children's Hospital of Eastern Ontario, Ottawa, Ontario, Canada K1H 8L1 Montreal Children's Hospital, Montreal, Quebec, Canada H3H 1P3 c British Columbia Children's Hospital, Vancouver, British Columbia, Canada V6H 3V4 d Hopital Ste-Justine, Montreal, Quebec, Canada H3T 1C5 e IWK Health Centre, Halifax, Nova Scotia, Canada B3K 6R8 f Hospital for Sick Children, Toronto, Ontario, Canada M5G 1X8 g Mt. Sinai Hospital, Toronto, Ontario, Canada M5G 1X5 b
Received 20 February 2012; accepted 5 March 2012
Key words: Gastroschisis; Bowel score; Outcomes; Prediction; Multivariate analysis; Validation; Risk stratification
Abstract Background/Purpose: Disease-specific outcome predictors are required for gastroschisis. We derived and validated a gastroschisis prognostic score (GPS) based on bowel appearance after birth. Methods: Visual scoring of bowel matting, necrosis, atresia, and perforation generated a novel gastroschisis bowel injury score recorded in a national database. Reweighting of score components by regression analysis led to assessments of model calibration and goodness of fit. The GPS was validated in subsequent cases. Results: Records from 225 infants were used for model derivation. Only intestinal necrosis independently predicted mortality by regression analysis (odds ratio, 11.5; 95% confidence interval, 4.2-31.4). Model recalibration identified that a GPS of 4 or more predicted mortality in 75% of nonsurvivors and 99% of survivors (P = .0001). A GPS of 2 or more demonstrated significantly worse survival outcomes compared with scores of 0 or 1 (length of stay: P = .011, days to first enteral feed: P = .013, days on total parenteral nutrition: P = .006). Model validation with 184 new patients yielded continued high-quality discrimination of outcomes. The GPS demonstrated “near-perfect” interobserver reliability between 2 surgeons (κ ≥ 0.86).
Abbreviations: GPS, gastroschisis prognostic score; CAPSNet, Canadian Pediatric Surgery Network; DTPNs, days on total parenteral nutrition; DFEFs, days to first enteral feed; LOS, length of stay; SNAP-II, Score for Neonatal Acute Physiology version II. ☆
Financial disclosure: None. Conflicts of interest: None. ⁎ Corresponding author. Tel.: +1 514 412 4438; fax: +1 514 412 4289. E-mail address: [email protected] (P.S. Puligandla). ☆☆
0022-3468/$ – see front matter © 2012 Elsevier Inc. All rights reserved. doi:10.1016/j.jpedsurg.2012.03.010
1112
K.N. Cowan et al. Conclusions: The GPS allows the accurate and reliable identification of high-risk groups for mortality and morbidity based on bowel appearance at birth. This information can drive discussions regarding family counseling, resource allocation, and new therapies for these patients. © 2012 Elsevier Inc. All rights reserved.
Gastroschisis is a congenital anomaly in which the bowel and other abdominal contents herniate through a paraumbilical, anterior abdominal wall defect. Although most of these infants survive, many experience significant gastrointestinal dysfunction [1]. Because the diagnosis of gastroschisis is made antenatally in most cases, attempts have been made to correlate antenatal ultrasonography findings with clinical outcome. Despite initial promise [2], variables such as bowel thickness and bowel dilatation on antenatal ultrasound have not been shown to be reliable predictors of morbidity or mortality in recent studies [3-6]. Postnatal variables such as prematurity and low birth weight have shown better correlation [7-9]. Furthermore, the classification of infants as complex, that is, those with intestinal atresia, stenosis, perforation, necrosis, or volvulus, has also been useful in identifying those infants at higher risk for morbidity and mortality [9-11]. These analyses, however, are often limited to single institutions and, therefore, may reflect small case numbers collected over long periods, often without the standardization of risk or outcome variables. The Canadian Pediatric Surgery Network (CAPSNet) is a multidisciplinary group from 16 tertiary perinatal centers that collects prospective, population-based data on gastroschisis and congenital diaphragmatic hernia. A major objective of CAPSNet has been the development of prognostication tools for gastroschisis. To date, no study has attempted to correlate the gross appearance of the bowel after birth with clinical outcome. For this purpose, we developed a standardized, visually determined, gastroschisis bowel injury scoring tool. The aim of this study was to use the prospective CAPSNet database to evaluate the outcome of infants with gastroschisis using a novel bowel injury scoring tool.
1. Patients and methods 1.1. Study population We performed a retrospective cohort study in which cases of gastroschisis born between May 1, 2005, and June 30, 2008, were abstracted from the CAPSNet database (inception cohort—cohort 1). This included data from diagnosis (usually prenatal) to death or discharge from the participating CAPSNet center. After initial analysis and modeling, study findings were prospectively validated in a new cohort of gastroschisis cases born between July 1, 2008, and October 31, 2010 (validation cohort—cohort 2).
1.2. Data collection Data were collected as previously described [4]. Briefly, a trained research assistant at each participating CAPSNet center abstracted prenatal and postnatal data using a customized data entry program and a standardized manual of operations and definitions with built-in error checking. The coded data, stripped of patient identifiers, was then transmitted electronically to a centralized, secure database for cleaning and storing. This process was overseen by a study coordinator and a multidisciplinary, geographically representative, steering committee consisting of pediatric surgeons, a neonatologist, a maternal-fetal medicine specialist, and an epidemiologist.
1.3. Gastroschisis bowel injury score To readily evaluate bowel injury in patients with gastroschisis and to standardize its assessment, a novel bowel injury scoring tool was developed. The composite score was based on key features of intestinal injury noted by the treating surgeon within 6 hours of birth. The score's component variables included the following: (a) bowel matting (none, 0; mild, 1; or severe, 2), (b) bowel necrosis (absent, 0; focal, 1; or diffuse, 2), (c) bowel atresia (absent, 0; suspected, 1, or present, 2), and (d) bowel perforation (absent, 0, or present, 2). Consensus-based, photographic examples with detailed, written descriptions of each quantitative bowel injury attribute were posted on the Web site (www.capsnetwork.org) in an attempt to standardize bowel scoring among surgeons. The injury score assessment was performed at the time of surgical consultation and prospectively recorded with the time of assessment.
1.4. Data analysis Data for analysis were abstracted directly from the CAPSNet database and analyzed with SPSS 12 for Windows statistical software (SPSS, Chicago, IL) using the tests subsequently described. The data parameters collected included demographic and perinatal risk variables (birth weight, gestational age, neonatal illness severity score [Score for Neonatal Acute Physiology version II, or SNAP-II] [12], and the gastroschisis bowel injury score). Mortality and nonmortality outcomes were analyzed, including length of stay (LOS), days to first enteral feed (DFEFs), days on total parenteral nutrition (DTPNs), and complications. The recorded complications included abdominal compartment
The gastroschisis prognostic score syndrome requiring treatment, culture-proven bacteremia or surgical site infection, a need for any reoperation, and the presence of severe TPN cholestasis at discharge (conjugated bilirubin N80 μmol/L). Univariate analyses for the comparison of risk-stratified groups were conducted using χ2 tests, whereas the MannWhitney test was performed for nonparametric variables without normal distribution. Multivariate predictors of mortality were analyzed with logistic regression where the conformity between actual and predicted outcome was assessed by model calibration and “goodness of fit” tests. To this end, the Omnibus test was used to assess the overall model fit, with a P value less than .05 confirming that the overall model parameters predicted mortality [13]. The Hosmer-Lemeshow technique was used to assess the goodness of fit, or the presence of a difference between the actual and modeled outcomes, where a significant model fit produced a P N .05 [14]. The individual components of the gastroschisis bowel injury scoring tool (matting, perforation, atresia, necrosis), as well as other putative outcome predictors identified in the literature, including gestational age, birth weight, and SNAPII [7,8,15], were subject to univariate analysis to identify those variables that were predictive of mortality. Variables displaying continued outcome prediction by logistic regression analyses were subsequently reweighted to derive a new prognostic score, denoted the gastroschisis prognostic score (GPS). The GPS was then used in all subsequent analyses including the risk stratification for mortality and the morbidity outcomes described previously. The validation of the GPS as an outcome prediction tool was subsequently performed on the next cohort of patients entered into the database from July 2008 to October 2010. Interobserver variability for the GPS was assessed using a second surgeon (blinded to the first surgeon's score) who provided a separate bowel scoring assessment based on a convenience sample of patients from the combined cohort. After biostatistical consultation, it was deemed that a convenience sample of more than 20% of the combined cohort was sufficient to analyze interobserver variability using Cohen κ statistic [16,17]. All secondary assessments for this analysis occurred between 6 and 24 hours of birth. All data analysis, methods, and statistical assessments were reviewed and overseen by biostatisticians and clinical research specialists from the University of Western Ontario (London, Ontario, Canada) and McGill University (Montreal, Quebec, Canada).
2. Results From a total of 510 gastroschisis patients entered into the CAPSNet database between May 2005 and October 2010, 409 patients with complete bowel scoring data were used for this study. There were no differences in demographic data
1113 including gestational age, birth weight, and SNAP-II score, nor mortality or survival outcomes between these patients and the 101 patients for whom complete bowel scoring data were not available (data not shown). Cohort 1 (inception cohort) consisted of 225 patients accrued from May 2005 to June 2008, whereas cohort 2 (validation cohort) consisted of the next 184 patients accrued from July 2008 to October 2010. The demographic details for the 2 cohorts are presented in Table 1. The overall mean gestational age and birth weight for the combined patient population were 36 weeks and 2526 g, respectively. The mean SNAP-II score for these patients was 8.4 ± 11.7 (median, 5; range, 0-64), suggesting that these infants had mild acute physiologic derangements and illness by this measure [12,18]. Ninetyone percent of all infants registered in the CAPSNet database during this time were evaluated with the bowel score within 6 hours of birth and exhibited a mean composite gastroschisis bowel injury score of 1 ± 1 (median, 1; range, 0-8). The overall survival rate for these patients was 96.8%. The median LOS was 37 days (range, 0-400 days), whereas the median DFEFs and median DTPNs were 14 (range, 1-78) and 28 (range, 4-309) days, respectively. All complications, when taken together, occurred at an average rate of 0.6 ± 0.8 events per patient (median, 0; range, 0-5). First, a predictive outcome model for mortality and survival outcomes was derived using bowel scoring (see “Methods”—“Gastroschisis bowel injury score”) and outcomes data for cohort 1. Multivariate logistic regression using only the 4 components of the gastroschisis bowel injury score demonstrated a robust model for the prediction of mortality in cohort 1. Univariate assessment of the individual contributions of each of the score components, as well as other potential predictor variables, identified that matting, necrosis, gestational age, and SNAP were variables that either had a greater individual contribution or could potentially contribute to the prediction model. Thus, these variables were candidates for either reweighting and/or Table 1 CAPSNet demographics and selected outcomes of patients with gastroschisis Parameter
Cohort 1 (n = 225)
Cohort 2 (n = 184)
Male, n (%) Gestational age (wk) Birth weight (g) SNAP-II b6 h to bowel score assessment, n (%) Gastroschisis bowel injury score Survival, n (%) LOS (median) (d) DFEFs (median) DTPNs (median) Complications
126 (56.0) 36 ± 2.2 2512 ± 583 8.5 ± 11.8 202 (89.8) 1±1 217 (96.4) 37 (1-349) 14 (2-78) 28 (5-221) 0.5 ± 0.8
96 (52.2) 36 ± 1.5 2543 ± 482 8.3 ± 11.8 172 (93.5) 1±1 179 (97.3) 36 (0-400) 14 (1-75) 28 (4-309) 0.6 ± 0.8
Data are presented as either mean ± SD or median (range) values, unless otherwise indicated.
1114
K.N. Cowan et al.
addition to the preliminary model that consisted of only the 4 components of the bowel injury score (Table 2). Importantly, univariate analysis identified necrosis, either focal or diffuse, as the only variable with the bowel injury score able to independently predict mortality (P = .0001) with an odds ratio of 11.5 (95% confidence interval, 4.2-31.4), whereas patients with matting, mild and severe assessed in aggregate, showed a trend toward significance (P = .055). Next, the individual components of the gastroschisis bowel injury score were reweighted within the prediction model based on the multiple logistic regression analyses (data not shown). Ultimately, the bowel injury score was reweighted in favor of both necrosis (absent, 0, or present, 4 [focal or diffuse]) and severe matting (none, 0; mild, 1; or severe, 4), thereby deriving a new prognostic score, denoted the GPS (Table 3). This revision directly reflected the independent predictive ability of any type of necrosis (ie, focal or diffuse) as well as the potential predictive ability of advanced (“severe”) matting. When these reweighted elements were combined into a mortality prediction model, the correct outcome was predicted in 99% (215/217) of survivors and 75% (6/8) of nonsurvivors. Furthermore, the Omnibus (P = .0001) and Hosmer-Lemeshow tests (P = .96) confirmed excellent model calibration and discrimination (Table 4). Interestingly, multivariate regression using the gastroschisis bowel injury score components with the addition of either gestational age or birth weight did not improve correlation with outcome, whereas the addition of the SNAP-II score, in fact, weakened the model fit (data not shown). The GPS permitted the identification of patients at a higher risk for death, with infants having a GPS of 4 or more (n = 38) experiencing a mortality rate of 16% (Table 4). When the patients with a GPS less than 4 (n = 187) were compared to this high-risk group, a statistically and clinically significant increase in survival (99%) was noted. Consistent with the univariate analysis, a significant difference in gestational age was observed between high- and low-risk groups, with higher gestational ages being associated with survival, whereas none was present when comparing groups on the basis of either birth weight or SNAP-II (data not shown). Table 2 Mortality outcome prediction for the components of the gastroschisis bowel injury score and other potential mortality predictor variables Parameter
P
Odds ratio
95% CI
Atresia Perforation Matting Necrosis Gestational age Birth weight SNAP-II
.108 .121 .055 .0001 .051 .179 .073
2.12 2.46 1.35 11.47 0.80 0.99 1.04
0.85-5.32 0.79-7.68 0.50-3.63 4.20-31.35 0.63-1.00 0.99-1.00 0.99-1.09
Abbreviation: CI, confidence interval.
Table 3 the GPS
Weighting “severe matting” and “necrosis” to derive
Matting Atresia Perforation Necrosis
None (0) Absent (0) Absent (0) Absent (0)
Mild (1) Suspected (1)
Severe (4) Present (2) Present (2) Present (4)
Based on logistic regression analyses, severe matting and necrosis were reweighted to acknowledge their individual contributions to the GPS. These criteria were initially given a maximum score of 2 in the preliminary model.
To be clinically relevant as a prognostic tool, it was important for the GPS to discriminate mortality from nonmortality outcomes, and for this reason, a separate logistic regression derivation of morbidity predictors was not performed. Rather, we used the same weighted scoring of bowel injury features that were used to predict mortality and performed a sequential analysis of composite GPS values (≥1, 2, 3, etc). This process identified a GPS of 2 or more as the threshold for the prediction of morbidity outcomes in cohort 1 (n = 51) (Table 5). Although there was no significant difference in the complication rates for patients with a GPS of 2 or more (n = 51) and those with a GPS less than 2 (n = 174), patients with a GPS of 2 or more demonstrated both a statistically and clinically significant increase in morbidity with respect to median LOS (63 vs 36, P = .011), DFEFs (19 vs 14, P = .013), and DTPNs (41 vs 27, P = .006). Next, model validation was undertaken using cohort 2. The GPS performed well, with excellent prediction of mortality in the validation cohort (Table 4). Moreover, the comparison of adverse survival outcomes between patients with a GPS of 2 or more and those with a GPS less than 2 demonstrated significant differences in all morbidities compared, including the rate of complications, an outcome that had only shown a trend toward significance in the inception cohort (cohort 1) (Table 5). Similar differences in mortality and adverse survival outcomes were observed Table 4 Risk-stratified mortality outcome prediction by the GPS in the derivation (cohort 1) and validation (cohort 2) patient groups Parameter
GPS ≥ 4
GPS b 4
Cohort 1 (n = 225) Mortality, n (%) Prediction of correct outcome (%) Omnibus test (P value) Hosmer-Lemeshow (P value) Cohort 2 (n = 184) Mortality, n (%) Prediction of correct outcome (%) Omnibus test (P value) Hosmer-Lemeshow (P value)
n = 38 6 (15.8) ⁎ 97.3 .0001 .96 n = 38 4 (10.5) † 97.3 .001 .921
n = 187 2 (1.1)
⁎ P = .0001. † P = .001.
n = 146 1 (0.7)
The gastroschisis prognostic score
1115
Table 5 Risk-stratified morbidity outcome prediction by the GPS in the derivation (cohort 1) and validation (cohort 2) patient groups Parameter
GPS ≥ 2
GPS b 2
Cohort 1 (n = 225) LOS (median) (d) DFEFs (median) DTPNs (median) Complications Cohort 2 (n = 184) LOS (median) (d) DFEFs (median) DTPNs (median) Complications
n = 51 63 (18-349) 19 (2-71) 41 (12-221) 0.8 ± 1.1 n = 49 68 (18-392) 19 (1-62) 50 (16-301) 0.9 ± 0.9
n = 174 36 (1-256) 14 (3-78) 27 (5-172) 0.5 ± 0.7 n = 135 33 (0-400) 13 (2-75) 27 (4-309) 0.5 ± 0.7
P .011 .013 .006 .06 .01 .002 .013 .003
Data are presented as either mean ± SD or median (range) values.
when both cohorts were examined in aggregate, where a significant difference in complication rate was again observed (data not shown). The interobserver variability in bowel injury scoring by the treating surgeon was evaluated using a second, blinded surgeon's score. This second evaluation occurred between 6 and 24 hours of birth and was applied to a convenience sample of 131 patients (32%) within the combined cohort. “Near perfect agreement” between observers was noted in the identification of patients at increased risk for mortality (GPS ≥ 4) and those at increased risk of morbidity (GPS ≥ 2), with κ values of 0.86 and 0.91, respectively [16,17].
3. Discussion To date, the prognostication of outcomes in infants with gastroschisis has remained controversial, and in light of the considerable resources used by these patients [19], there is much to be gained from a better understanding and, possibly, prediction of their clinical course [11]. To this end, many different antenatal and postnatal variables have been investigated [2,3,5,6,10,11]. This is the first study in the literature, however, to examine and correlate bowel appearance at birth with clinical outcome in the largest, populationbased group of infants with gastroschisis currently in the literature. This validated tool enables the identification of gastroschisis patients that are at high risk for mortality and morbidity in the immediate postnatal period. Risk stratification is not a new concept in gastroschisis. Caniano et al [10] and Molik et al [11] both published reports on the need to categorize infants with gastroschisis as either high or low risk based on several characteristics including the presence of intestinal atresia, stenosis, perforation, necrosis, or volvulus. These criteria continue to be used to identify high-risk patients [6,9], and some of these findings form the critical elements of the GPS. However, these risk stratification systems lack validation, and their dependence on the
diagnosis of intestinal atresia or stenosis (which may be delayed) limits their predictive use in the immediate newborn period. By contrast, the GPS is a rapid, efficient, and accurate means of risk categorization with excellent reproducibility that can be completed at the bedside without the need for monitoring or laboratory data. It can be performed within hours of birth using a simple, standardized, Web-based scoring tool that evaluates key attributes of bowel injury, specifically intestinal matting, atresia, perforation, and necrosis. A composite score of 2 or more heralds an infant at greater risk for multifactorial morbidity and prolonged hospitalization. A score of 4 or more identifies an infant with an additional higher likelihood of mortality. Although necrosis independently predicts mortality, severe bowel matting or a combination of bowel perforation, suspected atresia, and/or mild matting also leads to patient stratification into either of the increased risk groups. In this way, outcome prognostication can be performed shortly after birth to provide reliable and clinically relevant information that can be used to counsel families and select babies for innovative therapies, based on the anticipation of a complicated postnatal course, as well as providing a common benchmark upon which to compare future therapeutic interventions. Recently, the CAPSNet database used SNAP-II scoring to predict mortality in patients with gastroschisis [15]. This assessment, however, still required 12 hours of intensive hemodynamic, respiratory, and laboratory surveillance to calculate. Owen et al [20], in a population-based cohort of 393 infants with gastroschisis from the United Kingdom, avoided using the SNAP-II score in the risk stratification of patients because of the “burden” of data collection associated with this method. They simply defined complex patients as those with “additional bowel damage” such as perforation or atresia. In addition to SNAP-II, additional indicators of gestational maturity and fetal growth have also been shown to have a negative impact on LOS and feeding parameters in patients with gastroschisis [7,8]. Interestingly, mortality prediction was not improved by combinations of gestational age, birth weight, and the SNAP-II score with the GPS vs the GPS alone. This finding may reflect the failure of these individual variables to enhance the predictive ability already encompassed by the GPS, a potentially stronger predictor of outcome. An examination of the individual risk attributes of SNAP-II suggests that its predictive ability is focused on the uncommon clinical scenario of the physiologically unstable infant with gastroschisis [15]. Indeed, the risk of mortality was increased only for those infants with SNAP-II scores greater than 28, a value far exceeding the average SNAP-II score of 8 observed in the patient population with gastroschisis used in this study [15]. The GPS, by contrast, is, thus, a more useful predictor in its ability to accurately discriminate outcome in the more typical context of infants who are stable from a cardiorespiratory perspective, yet have a bowel injury of variable severity. Because the GPS has superior performance in this broader patient cohort, its use is more universal.
1116 Although the GPS was found to be a robust predictor of mortality outcomes based on Omnibus and HosmerLemeshow testing, it may still have limitations. The overall mortality rate in this cohort was very low, with only 13 deaths noted in the entire patient population. With such a low event rate, the model was only able to predict 77% (10/13) of these deaths. It is possible that with a considerably larger data set, the model's ability to predict mortality could improve. Furthermore, our results only reflect survival to discharge from hospital because we lack long-term follow-up data for these patients, and it is conceivable that patients with a high GPS could have died after discharge. Thus, given the very high overall survival rate in gastroschisis, accurate mortality prediction appears to be considerably less important than the prediction of adverse survival outcomes. Perhaps, the greatest use of the GPS is its ability to predict clinically relevant outcomes such as LOS, DFEFs, DTPNs, and complication rates. Therefore, we sought to establish risk thresholds that would discriminate between mortality and nonmortality outcomes. Although a separate model derivation for morbidities was considered, the difficulty in selecting the most “clinically important” morbidity outcome, as well as the questionable use of a tool that requires separate scoring systems for morbidity and mortality prediction, caused us to use a single tool (initially derived against the outcome of mortality), with different “threshold” composite scores that provided excellent discrimination of outcomes. Because a GPS less than 2 predicted a “low risk” of adverse outcome and a GPS of 4 or more predicted a “high risk” of adverse outcome (including mortality), we wondered if an independent “intermediate risk” group might also exist. However, the numbers of intermediate-risk patients (ie, GPS of 2 or 3) in isolation were comparatively small and did not provide as clear a separation of outcomes as we had hoped (data not shown). It is very possible that ongoing patient recruitment and the addition of long-term outcome data could enable us to better define an intermediate-risk group for both short- and long-term morbidity. Nonetheless, the value and use of the GPS rest in its ability to provide rapid and clinically relevant prognostic information on morbidity that can be used to drive important discussions regarding family counseling, institutional resource allocation, and the use of novel therapies for this population of patients.
4. Conclusion The GPS is an easy, efficient, reproducible, robust, and sensitive method for outcome prediction in neonates with gastroschisis. Unique, clinically relevant risk groups for mortality and morbidity can be identified using this score within hours of their birth. This information can then assist in the planning of multidisciplinary management regimens for these patients while informing discussions with parents related to outcome expectation.
K.N. Cowan et al.
Acknowledgments This work was supported by the Canadian Institute of Health Research Team in Maternal-Infant Care Grant MOP69050. We are indebted to Dr Jamie Seabrook and Dr Elise Mok of the Children's Health Research Institute, University of Western Ontario, and The Montreal Children's Hospital Research Institute, McGill University, respectively, for their statistical support throughout the course of the study. We would also like to thank Jennifer Claydon and Alana Gaumont for their superb management of the CAPSNet database. For their contributions to the preparation of this manuscript, we would also like to recognize Drs Andreana Bütter and Robert Baird.
References [1] Langer JC, Longaker MT, Crombleholme TM, et al. Etiology of intestinal damage in gastroschisis. I: effects of amniotic fluid exposure and bowel constriction in a fetal lamb model. J Pediatr Surg 1989;24: 992-7. [2] Bond SJ, Harrison MR, Filly RA, et al. Severity of intestinal damage in gastroschisis: correlation with prenatal sonographic findings. J Pediatr Surg 1988;23:520-5. [3] Piper HG, Jaksic T. The impact of prenatal bowel dilation on clinical outcomes in neonates with gastroschisis. J Pediatr Surg 2006;41: 897-900. [4] Skarsgard ED, Claydon J, Bouchard S, et al. Canadian Pediatric Surgical Network: a population-based pediatric surgery network and database for analyzing surgical birth defects. The first 100 cases of gastroschisis. J Pediatr Surg 2008;43:30-4 [discussion 34]. [5] Badillo AT, Hedrick HL, Wilson RD, et al. Prenatal ultrasonographic gastrointestinal abnormalities in fetuses with gastroschisis do not correlate with postnatal outcomes. J Pediatr Surg 2008;43:647-53. [6] Davis RP, Treadwell MC, Drongowski RA, et al. Risk stratification in gastroschisis: can prenatal evaluation or early postnatal factors predict outcome? Pediatr Surg Int 2009;25:319-25. [7] Ergun O, Barksdale E, Ergun FS, et al. The timing of delivery of infants with gastroschisis influences outcome. J Pediatr Surg 2005;40: 424-8. [8] Puligandla PS, Janvier A, Flageole H, et al. The significance of intrauterine growth restriction is different from prematurity for the outcome of infants with gastroschisis. J Pediatr Surg 2004;39:1200-4. [9] Payne NR, Pfleghaar K, Assel B, et al. Predicting the outcome of newborns with gastroschisis. J Pediatr Surg 2009;44:918-23. [10] Caniano DA, Brokaw B, Ginn-Pease ME. An individualized approach to the management of gastroschisis. J Pediatr Surg 1990;25:297-300. [11] Molik KA, Gingalewski CA, West KW, et al. Gastroschisis: a plea for risk categorization. J Pediatr Surg 2001;36:51-5. [12] Richardson DK, Corcoran JD, Escobar GJ, et al. SNAP-II and SNAPPE-II: simplified newborn illness severity and mortality risk scores. J Pediatr 2001;138:92-100. [13] Pearson ES, D'Agostino RB, Bowman KO. Test for departure from normality. Biometrika 1977;64:231-46. [14] Lemeshow S, Hosmer Jr DW. A review of goodness of fit statistics for use in the development of logistic regression models. Am J Epidemiol 1982;115:92-106. [15] Mills J, MacNab YC, Skarsgard ED, et al. Perinatal predictors of outcome in gastroschisis. J Perinatol 2010;30:809-13. [16] Cohen J. A coefficient of agreement for nominal scales. Educ Psychol Meas 1960;20:37-46.
The gastroschisis prognostic score [17] Viera AJ, Garrett JM. Understanding interobserver agreement: the kappa statistic. Fam Med 2005;37:360-3. [18] Chien LY, Whyte R, Thiessen P, et al. SNAP-II predicts severe intraventricular hemorrhage and chronic lung disease in the neonatal intensive care unit. J Perinatol 2002;22:26-30.
1117 [19] Sydorak RM, Nijagal A, Sbragia L, et al. Gastroschisis: small hole, big cost. J Pediatr Surg 2002;37:1669-72. [20] Owen A, Marven S, Johnson P, et al. Gastroschisis: a national cohort study to describe contemporary surgical strategies and outcomes. J Pediatr Surg 2010;45:1808-16.